SuttaCentral

SuttaCentral Voice v1.0.0 Released 😍

sc-voice
Tags: #<Tag:0x00007fc457a92c48>

#142

Thank you!

I’ve seen you’ve added it for v1.1.1. So far I’ve added my findings to the backlog or the mispronunciations list. Where is the best place to put them?


#143

Yeah, ideally, I think the mispronunciations list is a good place for them, but at the same time I’ll really believe in Bhante’s “put the shoe rake where the shoes go” principle. It doesn’t seem be be used so at least for now, just add 'em to the backlog and maybe we can try to get a bit more organised later on.


#144

Isn’t there three people as there are three voices.
Sorry for my dumb question.


#145

All Pali is Spoken by Adita.

The other 3 names (Amy, Raveena and Russell are also on the Amazon Polly list I linked above).


#146

I thought he is a Sri Lankan for some silly reason.
:rofl:


#147

So maybe better I move what I’ve put there to the backlog? Otherwise it might just be forgotten…


#148

Yes exactly. :slight_smile: :anjal:


#149

Done :white_check_mark: (plus a few more).


#150

AI voices are currently trained by voice samples from real people. For example, there really is a Siri voice actress who provided the original training samples. It is an extraordinary accomplishment to be a AI voice actor. One needs a warm, clear, pleasant voice that lacks idiosyncracies and is almost anonymous. Once trained, AI voices can generally speak new text with ease. However, speaking the suttas presents challenges unanticipated by the AI voice trainers.

Thanks to Sabbamitta, Bhante Sujato, Aminah, SCMatt and many others here at SuttaCentral, we have customized versions of the AWS AI voices that can have been taught to speak Pali. Aditi is an AI voice trained for bilingual English/Hindi, and has the best phonemes for Pali. I think that one day the technology will advance to the point where anyone might train an AI voice. When that happens, it would be possible to train an AI voice for Pali using voice samples provided by bhikkhus and bhikkhunis chanting in deep samadhi. For now we have Aditi, Amy, Raveena and Russell. And we shall be able to add more AI voices over the coming years for languages other than English.

Karl is an old Chinese guy who can’t speak Chinese. His mom did try to teach him Chinese, but he failed. Now he just goes for walks trying to recite DN33 with Amy and Aditi. That’s going better.


#151

Thanks, Karl
The work you have done is extraordinary.
SCV project take the Sutta reading to a whole new level.
For instance I have listend to five DN suttas since you upload the SCV.
:anjal:


#152

@Aminah, I think that adding SSL may be a critical feature of the next release. Without it, we will not be able to use the new subdomain voice.suttacentral.net. It’s also apparently going to be a bit of work. I spent several hours this evening trying to make it happen with a dummy SSL certificate with no luck. Let me know if I should work on other stuff next week.


#153

I have this problem with my new computer. Windows 10 running on Thinkpad.
SCV run perfectly on my old cheap computer!
Well, you don’t get what you paid for all the time.


#154

Absolutely, this is a drop everything-else-highest-priority.


#155

Hey @Aminah, thanks for nicely tidying up the RP, looks very clear now! :heart: :sunflower:


#156

I am listening to the DN6/7/8 now.
The bell sound ring in between halfway or attempt to ring between downloads.


#157

Thanks for reporting. This is also on the list already. (It also happens when listening to the sutta directly, without download.)


#158

Or actually, maybe please clarify more precisely: Do you mean in-between two suttas of a playlist of several suttas, or is it within the same sutta, in-between different sections? (The second is what I experienced.)


#159

It is withing the same Sutta and in-between different sections.


#160

OK, thanks. That’s the same we have already on the list then.


#161

And thanks to you for adding! :pray:

I can very well see the attraction to keeping things to a simple single list as has been Karl’s preference.

One of the things to try and keep on top of is setting out a fair workload in any one stretch. In terms of pronunciation items I wonder how you, Ang. @sabbamitta, and @karl_lew feel about this (at least interim) plan:

Ang. Sabbamitter, you maintain the main backlog pronunciation list prioritized as you see fit which I can then pull in one by one from the top into any single release plan. In any one release, I’ll stick to a max of three unchecked pronunciation issues (as one item gets checked I can add another after checking Karl’s capacity for the remainder of a given dev cycle).